Context engineering

Context engineering is the work of giving an AI analytics agent the structured knowledge it needs to answer questions correctly. That knowledge includes the database schema, the metric definitions, the business rules, the documentation, and the tools the agent can call.

It is the difference between a demo and an agent a team can trust. A model can write SQL on its own. It cannot know that your company excludes refunds from revenue, or that two tables that look alike serve different teams. Context engineering puts that knowledge in front of the agent in a form it can use every time.

Why it matters

Most analytics agents do not fail because the model is weak. They fail because the context is thin. When an agent has to guess what a metric means, it guesses confidently and is sometimes wrong, which is the worst outcome for anyone making a decision off the number.

Good context engineering treats this knowledge like code. It is written down, version controlled, tested, and governed, so the agent’s behavior is something you build on purpose rather than hope for.

How it works in practice

A data team building context usually connects the warehouse, syncs the schema, defines the core metrics in one place, writes down the rules that are not obvious from the data, and adds tests that check the agent against known questions. Open-source tools like nao are built around this exact workflow.

To see where context fits in the bigger picture, read how agentic analytics works.